ivangtorre/wav2vec2-xlsr-300m-quechua

This model was finetuned from a Wav2vec2.0 XLS-R model: 300M with the Quechua train parition of the Americas NLP 2022 dataset. This challenge took place during NeurIPSS 2022.

Example of usage

The model can be used directly (without a language model) as follows:

from transformers import Wav2Vec2Processor, Wav2Vec2ForCTC
import torch
import torchaudio

# load model and processor
processor = Wav2Vec2Processor.from_pretrained("ivangtorre/wav2vec2-xlsr-300m-quechua")
model = Wav2Vec2ForCTC.from_pretrained("ivangtorre/wav2vec2-xlsr-300m-quechua")

# Pat to wav file
pathfile = "/path/to/wavfile"

# Load and normalize the file
wav, curr_sample_rate = sf.read(pathfile, dtype="float32")
feats = torch.from_numpy(wav).float()
with torch.no_grad():
    feats = F.layer_norm(feats, feats.shape)
feats = torch.unsqueeze(feats, 0)
logits = model(feats).logits

# take argmax and decode
predicted_ids = torch.argmax(logits, dim=-1)
transcription = processor.batch_decode(predicted_ids)
print("HF prediction: ", transcription)

This code snipnet shows how to Evaluate the wav2vec2-xlsr-300m-quechua in Second Americas NLP 2022 Quechua dev set

from datasets import load_dataset
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
import torch
from jiwer import cer
import torch.nn.functional as F
from datasets import load_dataset
import soundfile as sf

americasnlp = load_dataset("ivangtorre/second_americas_nlp_2022", "quechua", split="dev")
quechua = americasnlp.filter(lambda language: language['subset']=='quechua')

model = Wav2Vec2ForCTC.from_pretrained("ivangtorre/wav2vec2-xlsr-300m-quechua")
processor = Wav2Vec2Processor.from_pretrained("ivangtorre/wav2vec2-xlsr-300m-quechua")

def map_to_pred(batch):
    wav = batch["audio"][0]["array"]
    feats = torch.from_numpy(wav).float()
    feats = F.layer_norm(feats, feats.shape) # Normalization performed during finetuning
    feats = torch.unsqueeze(feats, 0)
    logits = model(feats).logits
    predicted_ids = torch.argmax(logits, dim=-1)
    batch["transcription"] = processor.batch_decode(predicted_ids)
    return batch

result = quechua.map(map_to_pred, batched=True, batch_size=1)

print("CER:", cer(result["source_processed"], result["transcription"]))

Citation

@article{romero2024automatic,
  title={Automatic Speech Recognition Advancements for Indigenous Languages of the Americas},
  author={Romero, Monica and G{\'o}mez-Canaval, Sandra and Torre, Ivan G},
  journal={Applied Sciences},
  volume={14},
  number={15},
  pages={6497},
  year={2024},
  publisher={MDPI}
}